System and methods for customer quality prediction

Abstract

Apparatus and associated methods relate to determining scores rating historical customer quality, training a predictive analytic model to recognize historical customer quality determined as a function of ranking the scores, and predicting future customer quality based on the model. In an illustrative example, quality may be a vector quantity representing multi-source data. In some examples, the predictive analytic model may be trained to recognize a historical customer as a member of a subset of customers. For example, the model may be trained to recognize a customer subset selected based on a quality threshold characterizing the subset as good. In various embodiments, the predictive analytic model may be a neural network, permitting prediction based on weights adapted by machine learning techniques to learn which data sources are optimal predictors. Various examples may advantageously predict a customer quality trend as a function of time, permitting decisions based on predicted future customer quality.

What is claimed is: 
     
         1 . A process to determine customer value, the process comprising:
 determining scores rating historical customer quality;   training a predictive analytic model to recognize historical customer quality determined as a function of ranking the scores; and,   predicting future customer quality based on the model.   
     
     
         2 . The process of  claim 1 , wherein the scores further comprise multi-source data. 
     
     
         3 . The process of  claim 1 , wherein customer quality further comprises a vector quantity. 
     
     
         4 . The process of  claim 1 , wherein rating historical customer quality further comprises each score determined as a combination of multiple data components. 
     
     
         5 . The process of  claim 1 , wherein training the predictive analytic model further comprises partitioning historical data into train and test partitions. 
     
     
         6 . The process of  claim 1 , wherein training the predictive analytic model further comprises cross validation. 
     
     
         7 . The process of  claim 1 , wherein training the predictive analytic model further comprises a test to determine if the model prediction error is less than a predetermined maximum. 
     
     
         8 . The process of  claim 1 , wherein training the predictive analytic model further comprises a test to determine if the model is overfit based on cross-validation. 
     
     
         9 . The process of  claim 1 , wherein determining scores further comprises normalizing data. 
     
     
         10 . A process to determine customer value, the process comprising:
 determining scores rating historical customer quality, wherein the scores are determined as a function of multi-source data, and wherein customer quality is a vector quantity determined as a function of a plurality of multi-source data components;   training a predictive analytic model to recognize historical customer quality determined as a function of ranking the scores, wherein training the predictive analytic model includes partitioning historical data into train and test partitions, and, training the predictive analytic model using the train partition until the model prediction error satisfies a predetermined threshold;   testing the predictive analytic model, wherein testing the model includes cross-validation based on the test partition; and,   predicting future customer quality based on the model.   
     
     
         11 . The process of  claim 10 , wherein the predictive analytic model further comprises a neural network. 
     
     
         12 . The process of  claim 10 , wherein the predictive analytic model further comprises a random forest. 
     
     
         13 . The process of  claim 10 , wherein the predictive analytic model further comprises a decision tree. 
     
     
         14 . The process of  claim 10 , wherein the predictive analytic model further comprises a Bayesian classifier. 
     
     
         15 . The process of  claim 10 , wherein the train partition further comprises labeled data. 
     
     
         16 . The process of  claim 10 , wherein ranking customer quality further comprises evaluating an angle determined by vector analysis of data components. 
     
     
         17 . A process to determine customer value, the process comprising:
 determining scores rating historical customer quality, wherein the scores are determined as a function of multi-source data components, wherein customer quality is a vector quantity determined as a function of a plurality of the multi-source data components;   training a predictive analytic model comprising a neural network to recognize historical customer quality determined as a function of ranking the scores, wherein training the predictive analytic model includes partitioning historical data into train and test partitions, and training the predictive analytic model using the train partition until the model prediction error satisfies a predetermined threshold, and wherein ranking the scores further comprises evaluating an angle determined by vector analysis of data components;   testing the predictive analytic model, wherein testing the model includes cross-validation based on the test partition; and,   predicting future customer quality based on the model, wherein future customer quality is determined as a trend based on predicted quality evaluated for at least two points in time.   
     
     
         18 . The process of  claim 17 , wherein the future customer quality trend is determined as a function of the slope of a quality prediction line. 
     
     
         19 . The process of  claim 17 , wherein the future customer quality trend is determined as a function of CQIR. 
     
     
         20 . The process of  claim 17 , wherein customer quality is determined as a function of input from a property management system based on image data.